Announcement

Collapse
No announcement yet.

Luc Verhaegen Comments On Intel/Mir Politics

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #46
    Originally posted by johnc View Post
    X is the wave of the future.

    But putting that fact aside for the moment, is it pretty much certain now that there will be no -- zero, zip, nada -- Mir support in Intel drivers? Because it's going to be hard to jive the idea of not accepting an XMir patch on the basis of Intel "not condoning the actions of Canonical [by creating Mir]", and then separately providing support for Mir.
    Well, not condoning such actions might involve just XMir, if the action not condoned is using a compatibility layer for running the desktop. And that's as far as the ambiguity goes. It's likely that it refers to Mir as a whole, though.

    Also, there will be support, but until Intel says so, this support will be downstream, provided by Canonical.

    Comment


    • #47
      Originally posted by duby229 View Post
      Hold on here, How is it that you can blame your decisions on someone else? It was you that decided you didnt want to use Atombios, it was your decision to bang modesetting hardware directly. That had nothing to do with anyone else. If you want to place blame, then place it where it belongs. (You had all the documentation you needed, you just wanted documentation that was unreasonable to expect to get.)
      We had _a_ _lot_ of documentation for some things. We never got answers on the tough questions (spread spectrum, power management, bios re-reading after first driver start, tv-out, ...), or if we did, we got bogus information (The PLL wall: "why don't you run into the wall from another angle?"). AtomBIOS provided us quickly with the information we needed which was not documented in the register docs that everyone got, and it was quick and easy to implement those bits in C. All the bits of information which were not readily provided by the register docs, and which were not explained by AtomBIOS either, they really ate our time. And we would have to spend an hour every thursday listening why we didn't get an answer on this.

      Originally posted by duby229 View Post
      You chose to implement a development model that wasnt supportable. The reason that radeon surpassed radeonhd was because other developers could implement support for functionality and features that would have been far more difficult to do on radeonhd because of the choices you made.
      Did you know that ATI tried to block the release of RadeonHD by not allowing us to properly license the atombios interpreter code? Did you know that, because we only had limited dependence on atombios (we could make do with int10 booting and a manual description of the board layout), we stated: "fine, we ship without atombios then". This is why any code went out to begin with.

      Are you aware of what happened internally at AMD? Redhat was really unhappy at SuSE getting the scoop with radeonHD (hey, they should've helped AMD with the Hammer, but instead they couldn't be arsed) In Oktober 2007, the AMD account manager responsible for SuSE got moved aside, and the account manager responsible for Red Hat (Ted Donnely) took over this role. This is when the whole thing derailled. Mr Bridgman suddenly was free to play any games he liked, and he started supporting the competing project. In the end though, he played everyone against each other, making sure that fglrx survived as is.

      Did you also know that the endless corporate politics had forced us to implement more complete atombios dependence? Egbert was hard at work thoroughly designing and testing this. And then RV670 came round, and egbert, during the work on full atombios dependence, added 12 lines to the C code and had the hw work beautifully. Mr Bridgman was not happy with that.

      AtomBIOS was never the real issue, although it was not accepted by ATI politically. As soon as egbert added atombios dependent modesetting, the next stick to bash us with was found.

      Originally posted by duby229 View Post
      EDIT: I remember it clearly. It was mid 2007, radeonhd was still working on getting basic mode setting working and radeon implemented Atombios and supported more hardware than radeonhd instantly. There was no chance at all that you could provide the same level of support doing what you were doing. Additionally radeon was able to leverage work done for older generations to get a jump start for r600, which wasnt even a possibility on radeonhd. And radeon had more developers contributing to it.
      You seem to have your dates off. RadeonHD was released september 2007. Radeon forked the radeonhd code late november 2007.

      Yes, we were spending a lot of time on modesetting at radeonHD. Not on the atombios bits, but on the hard parts which no-one was able to tell us. Due to our extremely clean and transparent code, it was very easy to take the hard bits and use them with atombios and then calling it ones own code. And this is part of what made radeonhd seem slow. We were doing the hard bits and making none of the big noisy statements.

      Also, can you imagine the flack i would've gotten from the likes of airlied, ajax and daniels, if i, in 2007, had chosen to support atombios extensively? Intel only just dropped BIOS dependence. This would not have improved things in the radeonhd versus radeon battle either, as then radeon would've gone the proper C code way. It's not as if ATI would've been more supportive of us then, as ATI already was out to kill us after we had forced things open. It was a Catch-22 all along, and i picked the technically and morally more acceptable route.

      Radeon was playing with all sorts of things (r500 colourspace conversion, tv-out -- can you tell me what happened to the tv-out code btw? -- or was that really a case of airlied loudly shouting "it worked on my machine once! Absolute victory over radeonhd!!!!!"), and playing is the right word for it, while happily taking the hard modesetting bits we at radeonhd had solved.

      R600 modesetting was supported from the get-go on radeonhd. Matthias Hopf brought up the first triangle late 2008, with loads of conflicting information coming from ATI, in the end it was pure manual labour. Without a triangle on the R6xx, there was also none of these things like 3d engine colour space conversion or 2d acceleration.

      What really hurt us at radeonhd was my absolute resolve to create the best code possible and to have things tested as well as possible. RadeonHD bitbanging code was made for eternity and was written in such a way that it could have lived through any infrastructure change possible. Code versus noise always has noise win, but this was not what killed us.

      The real killer was AMD losing grip over ATI, and with time it was less and less able to support us. With code and docs public already, ATI could not afford to go back on that completely, but they could limit the fglrx damage and documentation overhead, and that's exactly what they did.
      Last edited by libv; 09-11-2013, 03:00 AM.

      Comment


      • #48
        Originally posted by libv View Post
        We had _a_ _lot_ of documentation for some things. We never got answers on the tough questions (spread spectrum, power management, bios re-reading after first driver start, tv-out, ...), or if we did, we got bogus information (The PLL wall: "why don't you run into the wall from another angle?"). AtomBIOS provided us quickly with the information we needed which was not documented in the register docs that everyone got, and it was quick and easy to implement those bits in C. All the bits of information which were not readily provided by the register docs, and which were not explained by AtomBIOS either, they really ate our time. And we would have to spend an hour every thursday listening why we didn't get an answer on this.



        Did you know that ATI tried to block the release of RadeonHD by not allowing us to properly license the atombios interpreter code? Did you know that, because we only had limited dependence on atombios (we could make do with int10 booting and a manual description of the board layout), we stated: "fine, we ship without atombios then". This is why any code went out to begin with.

        Are you aware of what happened internally at AMD? Redhat was really unhappy at SuSE getting the scoop with radeonHD (hey, they should've helped AMD with the Hammer, but instead they couldn't be arsed) In Oktober 2007, the AMD account manager responsible for SuSE got moved aside, and the account manager responsible for Red Hat (Ted Donnely) took over this role. This is when the whole thing derailled. Mr Bridgman suddenly was free to play any games he liked, and he started supporting the competing project. In the end though, he played everyone against each other, making sure that fglrx survived as is.

        Did you also know that the endless corporate politics had forced us to implement more complete atombios dependence? Egbert was hard at work thoroughly designing and testing this. And then RV670 came round, and egbert, during the work on full atombios dependence, added 12 lines to the C code and had the hw work beautifully. Mr Bridgman was not happy with that.

        AtomBIOS was never the real issue, although it was not accepted by ATI politically. As soon as egbert added atombios dependent modesetting, the next stick to bash us with was found.



        You seem to have your dates off. RadeonHD was released september 2007. Radeon forked the radeonhd code late november 2007.

        Yes, we were spending a lot of time on modesetting at radeonHD. Not on the atombios bits, but on the hard parts which no-one was able to tell us. Due to our extremely clean and transparent code, it was very easy to take the hard bits and use them with atombios and then calling it ones own code. And this is part of what made radeonhd seem slow. We were doing the hard bits and making none of the big noisy statements.

        Also, can you imagine the flack i would've gotten from the likes of airlied, ajax and daniels, if i, in 2007, had chosen to support atombios extensively? Intel only just dropped BIOS dependence. This would not have improved things in the radeonhd versus radeon battle either, as then radeon would've gone the proper C code way. It's not as if ATI would've been more supportive of us then, as ATI already was out to kill us after we had forced things open. It was a Catch-22 all along, and i picked the technically and morally more acceptable route.

        Radeon was playing with all sorts of things (r500 colourspace conversion, tv-out -- can you tell me what happened to the tv-out code btw? -- or was that really a case of airlied loudly shouting "it worked on my machine once! Absolute victory over radeonhd!!!!!"), and playing is the right word for it, while happily taking the hard modesetting bits we at radeonhd had solved.

        R600 modesetting was supported from the get-go on radeonhd. Matthias Hopf brought up the first triangle late 2008, with loads of conflicting information coming from ATI, in the end it was pure manual labour. Without a triangle on the R6xx, there was also none of these things like 3d engine colour space conversion or 2d acceleration.

        What really hurt us at radeonhd was my absolute resolve to create the best code possible and to have things tested as well as possible. RadeonHD bitbanging code was made for eternity and was written in such a way that it could have lived through any infrastructure change possible. Code versus noise always has noise win, but this was not what killed us.

        The real killer was AMD losing grip over ATI, and with time it was less and less able to support us. With code and docs public already, ATI could not afford to go back on that completely, but they could limit the fglrx damage and documentation overhead, and that's exactly what they did.
        Just get an NVIDIA card. Download driver. Done.

        Comment


        • #49
          Originally posted by F i L View Post
          To be fair (and i'm no X11 expert) X11 doesn't really have so many "bugs" as it has "architectural flaws".
          Well, I guess all the bugs have been ironed out throughout the millenia.

          Comment


          • #50
            Originally posted by libv View Post
            The real killer was AMD losing grip over ATI, and with time it was less and less able to support us. With code and docs public already, ATI could not afford to go back on that completely, but they could limit the fglrx damage and documentation overhead, and that's exactly what they did.
            I dont really know the tech details, and maybe radeonhd was better than radeon, for user experince it made no big difference which one u used. that matters for users. Another thing was that suse was bought by microsoft (yes thats not 100% true but not 100% false thats the problem). you cant accept money from the devil for promoting their os, and having linux patents and hope that you get love from the linux community. if there were over that additional intriges maybe not nice cant say much about that.

            But what I dont understand is why you all tell about ati and amd, was it amd or ati, there was a date xy where amd bought ati, after this time ati in juristic way didnt exist anymore. And I remember clearly that at least official there was 0 opensource commitment except the old r300 driver from ati before amd bought them. So how does ati matter with the radeonhd driver? they did not exist anymore (blessed god the day amd bought this bastard-microsoft-nsa-sponsored-company ati).


            And another thing even when you are right, and radeonhd should have succeded and it had nothing to do with the microsoft-bribes and other stuff I would agree that you dont want such a company with power over such import software-part, how does that then relate to the mir shit, mir is clearly nothing better than wayland so maybe at this time radeonhd should have be the winner because it was better AND because it was earlier there... so wayland is better and earlier their so why should somebody pay developers to support this garbage other than canonical itself?

            Comment


            • #51
              Originally posted by libv View Post
              Also, can you imagine the flack i would've gotten from the likes of airlied, ajax and daniels, if i, in 2007, had chosen to support atombios extensively?
              Perhaps understandable given the amount of time you spent railing against modesetting using anything which even looked like a BIOS, and telling everyone who did that that they were killing freedom. Similar to the about-face on proprietary drivers, which you only went quiet on when you contracted to Nokia to develop a proprietary driver for a year.

              Also, funny you rail hard against radeon's alleged NIH of radeonhd, when radeonhd was itself NIHing Avivo, but developed in secret due to your employer's requirements. Ho hum.

              Comment


              • #52
                Stopped reading after the first two lines.

                Oakley Sunglasses

                Last edited by Ra698shida; 09-11-2013, 11:55 PM.

                Comment


                • #53
                  Originally posted by daniels View Post
                  Perhaps understandable given the amount of time you spent railing against modesetting using anything which even looked like a BIOS, and telling everyone who did that that they were killing freedom.
                  I'm soo sorry about that. Soo bloody sorry that everyone now has a working intel driver. Soo sorry about having structure in display drivers today. Soo amazingly sorry about putting words like mode and setting together.

                  Originally posted by daniels View Post
                  Similar to the about-face on proprietary drivers, which you only went quiet on when you contracted to Nokia to develop a proprietary driver for a year.
                  Now you really are showing that you have no clue of what Imre and I were working towards at Nokia (which was sadly cut short by Elop). Perhaps if you hadn't been constantly absent and had kept your job, you might have found out. Maybe then you also hadn't felt the need to act out a few months later.

                  But yes, i am burned on powerVR, but only on powerVR, and i currently think that that is a good thing. I think i have slightly redeemed myself since, and, on top of that, managed to keep many good folks from wasting their time on powerVR.

                  Since you are working for a consulting company nowadays, why don't you tell all the little boys and girls what hardware you are burnt on?

                  And i do not think that i went too quiet, despite the fact of being busy rewriting large portions of the PVR GPLed kernel driver.

                  Originally posted by daniels View Post
                  Also, funny you rail hard against radeon's alleged NIH of radeonhd, when radeonhd was itself NIHing Avivo, but developed in secret due to your employer's requirements. Ho hum.
                  The only thing avivo did was show the way forwards for AMD, and that's exactly how it should be remembered. The information it contained, from my brief look before we actually started working on RadeonHD, was limited. And in your infinite wisdom you had decided to GPL it, and AMD disallowed us to talk to anyone, so we were not able to use it. If you had MITed it, or AMD/ATI had been as lacks with us as it was with its own employees, we would've actually built avivo up completely. You only have yourself to blame here.

                  Would you have MITed avivo if I had kindly asked you, without me being allowed to tell you why? I rather doubt it, as you weren't even willing to give us a mailinglist at x.org later on.

                  But by all means, point out the bits in RadeonHD where we took code or fixes/workarounds from avivo.
                  Last edited by libv; 09-13-2013, 07:27 AM.

                  Comment


                  • #54
                    Originally posted by libv View Post
                    Matthias Hopf brought up the first triangle late 2008, with loads of conflicting information coming from ATI, in the end it was pure manual labour. Without a triangle on the R6xx, there was also none of these things like 3d engine colour space conversion or 2d acceleration.
                    I actually ended up being the first to get a triangle running on r6xx/r7xx hardware; it was rv770 actually. We got r6xx working a little while later. Matthias and I worked really hard on that; I don't want to deny him any credit. It was not loads of conflicting information it was was just the general pain of bringing up a new 3D engine that was completely different from all previous asics. Major hw changes are always more painful the incremental ones. There was not a nice clean programming guide for the 3D engine at the time. I put together the stuff we did eventually release.

                    Comment


                    • #55
                      Originally posted by libv View Post
                      Would you have MITed avivo if I had kindly asked you, without me being allowed to tell you why?
                      Yes. But no-one ever asked.

                      Originally posted by libv View Post
                      But by all means, point out the bits in RadeonHD where we took code or fixes/workarounds from avivo.
                      You do realise that's what NIH means, right?

                      Comment


                      • #56
                        Originally posted by Ra698shida View Post
                        Stopped reading after the first two lines.

                        Oakley Sunglasses
                        Wow, this must be a boring thread if even the spambots stop reading it...

                        Comment


                        • #57
                          Originally posted by daniels View Post
                          Yes. But no-one ever asked.
                          I really rather doubt it, given the lengths of crap you've been throwing. But feel free to take this up with Mr Bridgman and his friends. During the summer of 2007, he was happily blabbing around while we at SuSE got a serious whipping for something we didn't even cause.

                          Originally posted by daniels View Post
                          You do realise that's what NIH means, right?
                          There was nothing to reinvent from avivo. It was all pretty limited from where i sit and sat.
                          Last edited by libv; 09-18-2013, 05:54 AM.

                          Comment


                          • #58
                            Originally posted by agd5f View Post
                            I actually ended up being the first to get a triangle running on r6xx/r7xx hardware; it was rv770 actually. We got r6xx working a little while later. Matthias and I worked really hard on that; I don't want to deny him any credit. It was not loads of conflicting information it was was just the general pain of bringing up a new 3D engine that was completely different from all previous asics. Major hw changes are always more painful the incremental ones. There was not a nice clean programming guide for the 3D engine at the time. I put together the stuff we did eventually release.
                            Ok, sounds plausible, but i will confirm that with him later on.

                            In any case, radeonHD supported R6xx just fine, from the second that R6xx support actually existed. Whoever claimed different earlier really had no clue what he was talking about.

                            Comment

                            Working...
                            X